An agent architecture with adaptive and learning capability

Author

Hwang, Kao-Shing ; Hsu, Harry Chia-Hung ; Liu, Alan

Author_Institution

Dept. of Electr. Eng., Nat. Chiing Cheng Univ., Chia-Yi, Taiwan

Volume

3

fYear

2002

fDate

28-31 Oct. 2002

Firstpage

1741

Abstract

The decoder of AHC uses the BOXES algorithm to divide the input state variables into several regions (or boxes). The drawback of this division is that it strongly depends on an engineer´s expertise. In addition, the region of each box cannot be modified even when the working environment changes or the resolution is unsatisfactory. To solve this problem, we use the ART theory to improve the decoding mechanism of the AHC architecture to propose and ART-based AHC architecture. We use this architecture to construct three agents. Each agent can control the mobile robot individually, and they perform well in simulations.

Keywords

ART neural nets; adaptive systems; learning (artificial intelligence); mobile robots; software agents; AHC; ART theory; BOXES algorithin; autonomous agent architecture; mobile robot; reinforcement learning; Autonomous agents; Bismuth; Control system synthesis; Decoding; Force control; Learning; Mobile robots; State-space methods; Subspace constraints; Wheels;

fLanguage

English

Publisher

ieee

Conference_Titel

TENCON '02. Proceedings. 2002 IEEE Region 10 Conference on Computers, Communications, Control and Power Engineering

Print_ISBN

0-7803-7490-8

Type

conf

DOI

10.1109/TENCON.2002.1182671

Filename

1182671